Teamwork & EDI AI-Chatbot

The Concept

In late 2025, I was forced to take a humanities class focused on team collaboration. For the final project, we were challenged to create something that could teach the rest of the class how to promote Equity, Diversity, and Inclusion (EDI) within teams.

While most groups opted for presentations or videos, I suggested to my team (a group of six) that we build a custom AI Chatbot trained on high-quality resources. The chatbot would be a virtual coach for students. They agreed, and I took the lead on the technical implementation.

Data Collection & Training

Garbage in, garbage out. The first step was curating a high-quality dataset.

1. Research: Each team member sourced articles and research papers on EDI principles. We filtered these down to the top three best resources.
2. Web Scraping: I identified four authoritative websites on teamwork dynamics and added them to the training pool.
3. Class Integration: I summarized our entire semester's worth of class notes into a single text document to ensure the bot used words relevant to our course.
4. Condensing: I manually reviewed and condensed the documents to ensure the AI only referenced the highest quality information. ✨

Tech Stack & Build

I needed an infrastructure that handled training, hosting, and building efficiently. I settled on Botpress; it was a platform I never used before, but it looked robust for this specific use case.

Infrastructure

• Engine: Botpress (for NLP and Knowledge Base management)
• Frontend: React
• Hosting: Vercel
• Styling: Custom palette matching the university's colors (Orange, Green, Blue, Red).

Development Challenges

Integrating the Botpress API into a custom React frontend wasn't entirely smooth sailing.

The main issue was Session Management. The bot would often create a new session on page reload which made it "forget" the previous context or conversation history. It took a few hours of debugging the API integration and tweaking the refresh logic so messages persisted correctly and the UI updated in real-time.

The Logic

Surprisingly, the AI required very little fine-tuning on the backend once the data was clean. Though I did modify the system prompt to enforce a specific response pattern.

The Bot's 4-Step Response Algorithm:

1. Identify Concern: Acknowledges the specific student issue.
2. Connect to EDI: Links the issue to one of the pillars of Equity, Diversity, or Inclusion. Then, explains its impact on team performance.
3. Propose Action: Offers concrete, actionable solutions.
4. Follow Up: Asks a follow-up question to keep the user engaged.

The tone was set to be conversational, empathetic, and professional. Like a supportive coach.

The Result

When we presented the bot to the class, the reception was excellent. The bot was responsive, accurate, and genuinely helpful.

It was a solid learning experience in RAG (Retrieval-Augmented Generation) workflows. Botpress proved to be a powerful tool for quickly spinning up knowledge-base bots. I'd definitely use it again for similar use cases where quick deployment and ease of training are priorities.

You can try it out here: PolyÉDI

Thanks for reading!